In this post, I will investigate the effects that ad spending can have on the outcome of elections on a district by district level. Using my current model, I will incorporate data found on the FEC’s website to approximate ad spend on a campaign district level. Using how much money was spent will be my variable, though I am aware that some scholarship says that number of ads rather than money spent is also a good indicator. I will also run tests on that next week. I will also be using data from the 2018 midterm to build my predictive model.
Advertising and Election Outcomes
Current Scholarship
Political scientists Gregory A Huber and Kevin Arceneaux have noted in their research that most districts tend to be non-competitive. They also argue that “Advertising does a little to inform, next to nothing to mobilize, and a great deal to persuade potential voters.” A key takeaway from their work is how campaign advertising can do a great deal to persuade potential voters to candidates. Therefore, for undecided voters, it is highly likely that receiving more of a certain candidates ads more often than another will increase the likelihood of that voter to vote for the candidate which they saw more ads for.
This has pretty important implications and would be useful to include into our model and test whether or not the amount campaigns spend on ads and air space can affect our model’s prediction and accuracy for both 2018 and the upcoming 2022 election.
My Existing Model
My current model from Week 4 included incumbency, economic factors, and basic polling. Currently, the R squared for this model is at .73, a fair amount. I will plot the actual dem vote pct share for accurate comparison.
| Observations | 438 |
| Dependent variable | DemVotesMajorPercent |
| Type | OLS linear regression |
| F(3,434) | 393.44 |
| R² | 0.73 |
| Adj. R² | 0.73 |
| Est. | S.E. | t val. | p | |
|---|---|---|---|---|
| (Intercept) | 79.51 | 3.89 | 20.43 | 0.00 |
| avg | -6.73 | 0.20 | -34.06 | 0.00 |
| Unemployed_prct | -0.10 | 0.98 | -0.11 | 0.92 |
| winner_candidate_incIncumbent | 3.28 | 1.28 | 2.56 | 0.01 |
| Standard errors: OLS |
My new local model that includes district level data on polling, incumbency, and local employment data is much more accurate than before. However, the R-squared of 0.73 which is the highest so far only when I use DemMajorVotePct as my outcome variable. When I do DemSeats(which my previous models used) as my outcome variable, I get a lower R-squared of 0.50 exactly.
Actual Outcomes from 2018
Below is a plot of the actual outcomes from the 2018 election from which im pulling my predictive modeling data from.
Updating the Model
Adding in Ad Spend as a Predictor
Now I have to add the variable of ad spending on a local level. I’ve gone ahead and downloaded data from the FEC for 2018 election spending data. This isn’t exactly the ad spend per campaign but I am using it as a proxy by making the assumption that the more money a particular race / candidate has overall translates to how much it is spending in ads.